Genetic algorithm for optimal imperceptibility in image communication through noisy channel Santi P. Maity 1 , Malay K. Kundu 2 and Prasanta K. Nandi 3 1 Bengal Engineering College (DU), P.O.-Botanic Garden, Howrah, India, 711 103 2 Machine Intelligence Unit, 203 B. T. Road, Kolkata, India, 700 108 3 Bengal Engineering College (DU), P.O.-Botanic Garden, Howrah, India, 711 103 Abstract. Data embedding in digital images involves a trade off rela- tionship among imperceptibility, robustness, data security and embed- ding rate etc. Genetic Algorithms (GA) can be used to achieve optimal solution in this multidimensional nonlinear problem of conflicting na- ture. The use of the tool has been explored very little in this topic of research. The current paper attempts to use GA for finding out values of parameters, namely reference amplitude (A) and modulation index (μ) both with linear and non linear transformation functions, for achieving the optimal data imperceptibility. Results on security for the embedded data and robustness against linear, non linear filtering, noise addition, and lossy compression are reported here for some benchmark images. 1 Introduction The properties of data hiding techniques in digital images such as perceptual transparency, higher capacity, statistical invisibility or security of the hidden data, and robustness to some types of attacks are related in a conflicting manner and the design tradeoff depends on the applications (see [1]and [2]). Genetic Algorithms (GAs) can be used to optimize the conflicting requirements of data hiding problem but the use of the tool has been explored very little in this topic of research although there are many problems in the area of pattern recognition and image processing [3] where GAs perform an efficient search in complex spaces in order to achieve an optimal solution. In the present paper GA is used to find two parameter values, namely ref- erence amplitude (A) and modulation index (μ) in linear and non linear trans- formation functions used to modulate the auxiliary message. The cover object (image) is chosen in which the message is well hidden unlike in watermarking applications where the message must be hidden in the cover work to which it refers and no other. Image regions with relatively low information content and having pixel values in the lower and upper portion of the dynamic range are used for data hiding as the characteristics of human visual system (HVS) are less sensitive to the change at these two ends. The paper is organized as follows. Sections 2 describes transformation func- tions for message modulation and how to calculate the range of parameters. Genetic algorithm for data embedding is presented in section 3 with embed-